Tool identification method, tool identification device, adjustment and / or measurement apparatus, and computer-implemented method

Abstract

The invention relates to a tool identification method at least for recognising tools (10) which are designed as cutting tools, as profile machining tools and / or as forming cutters, the method comprising: at least one detection step (12), in which at least one transmitted light image (14) at least of the tool (10) is recorded; and at least one recognition step (16), in which at least one tool type of the tool (10) is determined by means of a trained, and in particular further trainable, algorithm, in particular an AI algorithm, preferably a machine learning algorithm, at least on the basis of the transmitted light image (14) recorded in the detection step (12).

Description

[0001] Tool identification method, tool identification device, setting and / or measuring device and computer-implemented method

[0002] State of the art

[0003] The invention relates to a tool identification method according to claim 1, a tool identification device according to claim 13, a setting and / or measuring device according to claim 14, a computer program product and / or a computer program computing infrastructure according to claim 15 and a computer-implemented method according to claim 16.

[0004] Cutting tools and / or form cutters are often recognized in measuring and setting devices by reading identifiers such as QR codes or ID chips or specified by individual user input.

[0005] The object of the invention is, in particular, to provide a generic method with advantageous properties regarding process reliability. This object is achieved according to the invention by the features of the independent patent claims, while advantageous embodiments and further developments of the invention can be found in the subclaims.

[0006] Advantages of the invention

[0007] A tool identification method is proposed at least for the recognition of tools designed as cutting tools, as profile machining tools and / or as form milling cutters, in particular for machining metal, stone and / or wood, with at least one detection step in which at least one transmitted light image of at least the tool is recorded, and with at least one recognition step in which at least one tool type of the tool is determined by means of a trained and in particular further trainable algorithm, in particular a KI algorithm, preferably a machine learning algorithm, at least based on the transmitted light image recorded in the detection step. This makes it possible to achieve advantageous properties with regard to process reliability. The error detection rate of tools can advantageously be reduced. Advantageously, e.g.The error-prone use of identifiers such as ID chips or QR codes, which are prone to interchangeability or damage, can be avoided. Advantageously, for example, in a setting and measuring device or in a machine tool, the selection of stored tool data can be simplified and made less prone to errors, particularly by significantly reducing the risk of using incorrect or inappropriate tool data. Furthermore, time and costs can be reduced, for example, by eliminating the need for systems for reading and transmitting the aforementioned identifiers.

[0008] A cutting tool is to be understood, in particular, as a tool for a metal-cutting / machining manufacturing process. A cutting tool preferably comprises at least three parts: a shaft, a handle in the case of manual tools or a machine interface in the case of machine tools, and a part that is effective during machining, such as a cutting part. The cutting part of a cutting tool penetrates into a material of the workpiece, in particular during machining of a workpiece. For example, the cutting tool can be designed as a turning tool, a milling cutter, a drill, a countersink drill, a reamer, a saw, a broaching tool, a scraper, a plane chisel, a brush, a file, or a rasp. In particular, the cutting tool is a machine tool cutting tool for use in a machine tool, in particular an industrial one.The cutting tool is preferably designed differently from a surgical tool. The profile machining tool can be designed, for example, as a profile polishing wheel or as a profiling wheel, in particular for machining side edges of stone slabs or wooden panels. In particular, the profile machining tool, preferably the profiling wheel and / or the profile polishing wheel, has a machining surface, preferably a milling surface or a polishing surface, which extends around a circumference of the profile machining tool, in particular around a tool rotation axis of the profile machining tool. In particular, the machining surface, preferably the milling surface or the polishing surface, forms a shape, in particular the working contour, which at least substantially corresponds to the desired edge shape of the stone slab or wooden panel to be machined.The stone slab can be designed, in particular, as a kitchen stone slab or as a vanity stone slab or the like. The wood slab can be designed as a furniture wood slab. A form cutter is understood, in particular, to be a milling tool (i.e., a cutting tool) that does not contain the shape of a workpiece to be produced. Form cutters are preferably intended for use in CNC-controlled milling machines. For example, the form cutter can be designed as a lubrication groove cutter, an angle face cutter, an angle face roughing cutter, an angle face end mill, a prism cutter, a semicircular form cutter (concave or convex), a quarter circle form cutter, an end rounding cutter, a worm cutter, a worm gear cutter (with spindle), a hob cutter, a sprocket cutter, a rack cutter, or a tooth form cutter.

[0009] The transmitted-light image is recorded, in particular, by a transmitted-light camera system. The transmitted-light camera system comprises, in particular, at least one camera and at least one illumination unit, which is arranged on a side of an object to be recorded by the transmitted-light camera system, for example, the tool to be recorded by the transmitted-light camera system, opposite the camera. The illumination unit of the transmitted-light camera system preferably generates particularly homogeneous and / or particularly parallel light. In particular, the transmitted-light image comprises a silhouette and / or a contour, in particular an enveloping contour, of the object to be recorded, in particular the tool. In particular, the transmitted-light camera system comprises only a single camera sensor. In particular, the tool identification method can be carried out without a stereo camera, without a tactile sensor, and / or without a laser sensor / laser scanning sensor.In particular, a tool identification device provided for carrying out the tool identification method is designed free of a stereo camera, free of a tactile sensor and / or free of a laser sensor / laser scanning sensor. In particular, no three-dimensional image data are required to carry out the tool identification method. This advantageously allows complexity to be kept low. The trained and / or trainable algorithm is provided in particular for a classification and / or regression of the input data, i.e. the transmitted light images. A "tool type" is to be understood as a set of tools that agree in essential properties. For example, the tool class of cutting tools includes the tool types prism milling cutters and angle face milling cutters, which in turn each comprise several tool models and / or tool series (e.g.Prismatic cutters of various sizes). In particular, the trained algorithm is trained using a computer-implemented method for training a machine learning algorithm, described below.

[0010] It is further proposed that the trained, and in particular further trainable, algorithm can determine at least one tool model / tool ​​series of tools of identical tool type in the recognition step based on the transmitted light image recorded in the acquisition step. This can advantageously increase process reliability even further. Advantageously, for example, a tool misrecognition rate can be reduced even further. If the trained and / or trainable algorithm is a CNN (Convolutional Neural Network) algorithm, advantages can be achieved in particular when processing larger amounts of data during image recognition based on the transmitted light images. In addition, advantages can also be achieved in image recognition of suboptimal transmitted light images that exhibit image distortions and / or different lighting conditions.In addition, the memory requirement can be advantageously kept low compared to other neural networks. For example, one of the well-known CNN algorithms described in the following publications can be used in the tool identification process: a) AlexNet: Alex Krizhevsky, Imagenet classification with deep convolutional neural networks, Communications of the ACM 60.6, pp. 84-90 (2017); b) MobileNet: Andrew G. Howard, MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications, CoRR, abs / 1704.04861 , (2017); c) Xception : Francois Chollet, Xception: Deep Learning with Depthwise Separable Convolutions, CoRR, abs / 1610.02357, (2016); d) LeCun Y, Bengio Y, Hinton G (2015) Deep learning; Nature 521 :436{444, DOI 10.1038 / naturel4539; e) Lin H, Li B, Wang X, Shu Y, Niu S (2019); Automated defect inspection of LED chip using deep convolutional neural network; J Intell Manuf; 30:2525{2534, DOI 10.1007 / sl0845-018-1415-x; f) Fu G, Sun P, Zhu W, Yang J, Cao Y, Yang MY, Cao Y (2019); A deep-learning-based approach for fast and robust steel surface defects classification; Opt Laser Eng 121 :397{405, DOI 10.1016 / j.optlaseng.2019.05.005; g) Lee KB, Cheon S, Kim CO (2017) A Convolutional Neural Network for Fault Classification and Diagnosis in Semiconductor Manufacturing Processes; IEEE T Semiconduct M 30: 135{142, DOI 10.1109 / TSM.2017.2676245; h) Goncalves DA, Stemmer MR, Pereira M (2020) A convolutional neural network approach on bead geometry estimation for a laser cladding system; Int J Adv Manuf Tech 106:1811 {1821 , DOI 10.1007 / s00170-019-04669-z; i) Karatas A, Kölsch D, Schmidt S, Eier M, Seewig J (2019) Development of a convolutional autoencoder using deep neuronal networks for defect detection and generating ideal references for cutting edges; Munich, Germany, DOI 10.1117 / 12.2525882; j) Stahl J, Jauch C (2019) Quick roughness evaluation of cut edges using a convolutional neural network; In: Proceedings SPIE 11172, Munich, Germany, DOI 10.1117 / 12.2519440; or k) a CNN from the open source framework known as “TensorFlow.” Alternative CNN algorithms known to those skilled in the art, such as Region Proposals (R-CNN, Fast R-CNN, Faster R-CNN), Detectron, Single Shot MultiBox Detector (SSD), or You Only Look Once (YOLO, for example, in version 8, which is available for licensing at the time of registration), etc., are of course also conceivable. Simple open-source solutions are available for applying many of these machine learning algorithms (see the article from the online encyclopedia “Wikipedia” on object recognition: https: / / en.wikipedia.org / wiki / Outline_of_object_recognition, as of: Revision of October 30, 2023 - 12:14 p.m.).In particular, the trained and / or trainable algorithm, in particular a Kl algorithm, preferably a machine learning algorithm, is executed by a computing unit, which can be part of a setting and / or measuring device or which can be arranged separately from a setting and / or measuring device. A “computing unit” should be understood in particular to be a unit with an information input, an information processing unit, and an information output. Advantageously, the computing unit has at least one processor, a memory, input and output means, further electrical components, an operating program, control routines, control routines and / or calculation routines. Preferably, the components of the computing unit are arranged on a common circuit board and / or advantageously arranged in a common housing. Alternatively, however, the computing unit can also be designed as a distributed computing unit, such as a cloud.In particular, the trained and / or trainable algorithm comprises a recognition algorithm and / or a classification algorithm, in particular an object classification algorithm.

[0011] Furthermore, it is proposed that in the recognition step, at least the tool type of the tool, preferably an exact tool model / a tool series of a tool, is determined by means of the trained, and in particular further trainable, algorithm at least based on an envelope contour of the tool determined, in particular by means of an envelope contour determination algorithm, from the transmitted light image recorded in the detection step. This can advantageously enable efficient and / or fast pattern recognition. The envelope contour of the tool is determined, in particular, from a silhouette of the tool recorded with the transmitted light image. The envelope contour can be determined from the transmitted light image by means of the computing unit, in particular by means of the envelope contour determination algorithm specifically provided for this purpose or by means of the trained, and in particular further trainable, algorithm from the detection step.“Intended” should be understood in particular to mean specially programmed, designed and / or equipped. The fact that an object is intended for a specific function should be understood in particular to mean that the object fulfills and / or performs this specific function in at least one application and / or operating state. The envelope contour is in particular a boundary line that completely surrounds an object shown in transmitted light. The envelope contour is in particular a two-dimensional contour. In particular, in the detection step, the envelope contour is first determined from the image data of the transmitted light image, for example using the envelope contour determination algorithm. Subsequently, the trained algorithm then determines at least the tool type of the tool, preferably an exact tool model / tool ​​series of a tool, from the envelope contour determined in this way.The envelope contours thus serve as, in particular the sole, input data for the trained algorithm. The envelope contour determination algorithm is preferably an envelope contour determination algorithm known to those skilled in the art, in particular an edge detection algorithm, such as a Canny edge detector algorithm, a Sobel operator algorithm, a Laplace filter algorithm, or the like. In particular, the envelope contour of an object within the meaning of this document is formed by a silhouette / outline of the object.It is further proposed that in the recognition step, at least the tool type of the tool, preferably an exact tool model / an exact tool series of a tool, is determined by means of the trained, and in particular further trainable, algorithm at least based on a tool position of the tool, in particular an absolute or relative one determined from the transmitted light image recorded in the detection step, in particular by means of a position recognition algorithm and / or by means of the trained, and in particular further trainable, algorithm. This advantageously makes it possible to make tool identification particularly precise. This advantageously enables, for example, a differentiation between tools that are merely different in length and otherwise of the same type.In particular, the position detection algorithm can be designed as a further trained, and in particular further trainable, algorithm, in particular a further AI algorithm, preferably a further machine learning algorithm, or as a classic (untrainable) algorithm. For example, the trained, and in particular further trainable, algorithm could determine a tool type, e.g. based on a cutting edge shape of a tool, in a sub-step of the detection step, while the position detection algorithm determines a dimension, e.g. a length, of the tool in a further sub-step of the detection step. The exact tool model / the exact tool series can then be derived from the tool type and the dimension when viewed together. The trained algorithm can preferably be further trained by an algorithm training step described in more detail below.Through further training, the accuracy of the trained algorithm should be continuously improved.

[0012] If the tool is positioned in a positioning step, in particular one which precedes the detection step, in particular at least partially automated, preferably fully automated, in a reproducibly definable, exactly ascertainable and / or exactly known relative position to a transmitted-light camera system recording the transmitted-light image and / or if the transmitted-light camera system recording the transmitted-light image is positioned in an alternative or additional positioning step, in particular one which precedes the detection step, in a reproducibly definable, exactly ascertainable and / or exactly known relative position to the tool, a particularly precise differentiation between tools which are merely of different lengths and otherwise of the same type can advantageously be enabled.Advantageously, a high level of user-friendliness can also be achieved, in particular through largely automatic tool identification. In particular, a motor-actuated tool positioning device is provided, which can move the tool at least translationally and / or rotationally to set the desired relative position. The actuated tool positioning device can be part of a setting and / or measuring device or can be designed separately from a setting and / or measuring device. In particular, a motor-actuated camera positioning device is provided, which can move at least components of the transmitted-light camera system or the entire transmitted-light camera system at least translationally and / or rotationally to set the desired relative position.The actuatable camera positioning device can be part of a setting and / or measuring device or can be designed separately from a setting and / or measuring device.

[0013] It is additionally proposed that the transmitted-light image recorded in the recording step comprises at least 40%, preferably at least 60%, and preferably 100%, of a machining contour, in particular a machining outline, of the tool, in particular a machining surface of the tool, in at least one side view of the tool. This advantageously makes it possible to achieve a high level of reliability and / or accuracy in tool identification. It is also conceivable that, in the recording step, several transmitted-light partial images of different partial regions of the machining contour are recorded and, in particular, pieced together to form a larger transmitted-light image, preferably by a so-called "stitching method," e.g., by means of the computing unit. In particular, the transmitted-light camera system has a camera sensor with a large detection range.The detection range of the camera sensor of the transmitted-light camera system has, in particular, at least the dimensions 50 mm x 50 mm, preferably 80 mm x 80 mm, preferably 100 mm x 100 mm, and particularly preferably 120 mm x 120 mm, especially with a working distance between the tool and the camera sensor of the transmitted-light camera system of approximately 120 mm. However, camera sensors with smaller dimensions are also possible.

[0014] Detection areas are conceivable, e.g., approximately 3.5 mm x 3.5 mm, approximately 6.5 mm x 5 mm, or approximately 14 mm x 12 mm. The machining contour can be formed, particularly depending on the tool, by a cutting contour, a milling contour, a grinding contour, or a polishing contour.

[0015] It is also proposed that the transmitted light image recorded in the detection step be high-resolution. This advantageously makes it possible to achieve a high level of reliability and / or accuracy in tool identification. This can advantageously be used, for example, to advantageously detect a grinding grit or the like. It is conceivable that the transmitted light image is examined using an algorithm, e.g. running on the computing unit, for detecting surface roughness and / or edge roughness. This algorithm could be provided to determine a root mean square roughness or a mean roughness from the recorded transmitted light image, e.g. from the grinding grit read out from the transmitted light image, from which an exact tool model / an exact tool series of a tool of a specific tool type can then preferably be determined.For example, the trained and / or trainable algorithm could determine a tool type in one sub-step of the acquisition step, e.g. based on a milling, grinding or polishing contour of a tool, while the algorithm for detecting the surface roughness and / or the edge roughness determines the root mean square roughness or the mean roughness in a further sub-step of the acquisition step. The exact tool model / the exact tool series can then be derived from the tool type and the root mean square roughness or the mean roughness in a synopsis. In particular, to accelerate the sub-step of the acquisition step in which the tool type is determined by the trained, and in particular further trainable, algorithm, the high-resolution transmitted light image can be reduced to a lower resolution, e.g.a resolution reduced, in particular in at least one dimension, preferably in two dimensions, by at least a factor of 2, preferably at least a factor of 4, preferably at least a factor of 6 and particularly preferably at least a factor of 10. In particular, a reduction by a factor of 10 in one dimension can mean a reduction in resolution in “pixels per area” (two dimensions) by a factor of 100. In particular, in order to accelerate the sub-step of the detection step in which the tool type is determined by the trained, and in particular further trainable, algorithm, the high-resolution transmitted light image is downscaled to a resolution which no longer allows detection of the roughness or the mean roughness.In particular, in the further sub-step of the detection step, in which the algorithm for detecting the surface roughness and / or edge roughness determines the square roughness or the mean roughness, the high-resolution transmitted light image is analyzed. Advantageously, by distributing the tasks among different algorithms, which analyze transmitted light images with different high resolutions, the total computing time and total computing effort can be kept particularly low. In particular, the camera of the transmitted light camera system has a resolution which, at a working distance between the tool and the camera of approximately 120 mm, allows the detection of shapes, objects or surface protrusions with a size of at most 70 pm, preferably of at most 50 pm, advantageously of at most 40 pm, preferably of at most 32 pm, and particularly preferably of at most 25 pm.In particular, the camera of the transmitted-light camera system has a spatial resolution of approximately 40 pm to 65 pm, preferably approximately 20 pm to 32 pm, at a working distance of 120 mm. In particular, the camera of the transmitted-light camera system has a resolution of at least 2768 x 1846 pixels, preferably at least 5536 x 3692 pixels. In particular, the camera of the transmitted-light camera system or the high-resolution transmitted-light image, preferably the transmitted-light image downscaled by one of the aforementioned factors, has a resolution of at least 0.5 megapixels (Mp), preferably at least 1 Mp, advantageously at least 5 Mp, preferably at least 12 Mp, and particularly preferably at least 20 Mp. It is conceivable that the camera of the transmitted-light camera system is read out with a binning of 2 x 2 to increase the capture rate, which in particular halves the maximum resolution in pixels.

[0016] Furthermore, it is proposed that the tool identification method include an output step in which the transmitted light image of the tool used to perform the recognition step and / or the determined tool type, preferably the determined tool model / the determined tool series, is output to a user either automatically or only upon prior request by the user. This advantageously further increases process reliability, for example, by further reducing the risk of incorrect tool recognition. Furthermore, a training function can advantageously be created to optimize the trainable algorithm, in particular the machine learning algorithm.

[0017] In this context, it is proposed that the tool identification method comprises the algorithm training step in which feedback from the user receiving the transmitted light image of the tool and / or the determined tool type of the tool, preferably the determined tool model / the determined tool series of the tool, in the output step regarding the correctness of an identification of the tool type of the tool, in particular of an exact tool model / an exact tool series of the tool, is used for automated optimization and / or for automated training (e.g. supervised learning of a neural network) of the trainable algorithm carrying out the recognition step, in particular of the Cl algorithm, preferably of the machine learning algorithm.This advantageously allows for continuous improvement of the accuracy and / or reliability of the tool identification method and / or enables adaptation to specific user constraints (the tool types primarily used). For example, in this case, insights gained through machine learning could be exchanged between networked tool identification devices. For example, in the algorithm training step, after receiving the transmitted light image and / or the determined tool type, preferably the determined tool model / tool ​​series, the receiving user confirms or rejects a tool classification, in particular the specific tool type, preferably the specific tool model / tool ​​series.In particular, the trainable algorithm takes previous confirmations or rejections into account when creating future tool classifications. It is conceivable that the algorithm training step takes place internally in the processing unit of the setting and / or measuring machine. However, the algorithm training step is preferably outsourced to an external processing unit, which is in contact with the transmitted-light camera system, for example, via a data transmission unit, particularly due to the computing power requirements.

[0018] If the transmitted light image used to carry out the recognition step is recorded of a stationary tool, a high speed of tool recognition can advantageously be achieved. Advantageously, a particularly uncomplicated tool identification method can be obtained, in particular with a particularly simple recording step, in particular in contrast to the use of dynamic transmitted light images, which are more complex at least in terms of recording time and are also considerably more complex to evaluate. In particular, a transmitted light image of the stationary tool is recorded in the recording step. In particular, the transmitted light image of the stationary tool is used to carry out the following recognition step. Preferably, only transmitted light images of stationary tools are used to carry out the recognition steps.Preferably, the trained algorithm is applied exclusively to transmitted light images of stationary tools. In particular, the trained algorithm is trained (exclusively) with transmitted light images of stationary tools. Preferably, the transmitted light images used and generated in the tool identification method are different from dynamic image recordings / dynamic transmitted light images, as shown, for example, in Chinese patent CN 112683193 B in Figures 2a to 2d and 3a. In particular, the transmitted light image is a transmitted light still image. In particular, the transmitted light image is a non-dynamic transmitted light image. In particular, the tool can be rotated into an optimal rest position before recording the transmitted light image, preferably the transmitted light still image, for example into a rest position in which a maximum transverse extent of the silhouette of an object contained in the transmitted light image is maximum.

[0019] Furthermore, the tool identification device, at least for recognizing tools designed as cutting tools, profile machining tools, and / or form milling cutters, in particular for carrying out the tool identification method, is proposed, comprising the transmitted-light camera system for recording at least one transmitted-light image of at least the tool, and comprising the internal or external computing unit configured to determine at least one tool type of the tool, and preferably an exact tool model / an exact tool series of the tool, by means of the trained, and in particular further trainable, algorithm, in particular a K1 algorithm, preferably a machine learning algorithm, at least based on the transmitted-light image of the transmitted-light camera system. Advantageous properties with regard to process reliability can thereby be achieved.Advantageously, the error detection rate of tools can be reduced.

[0020] Additionally, the setting and / or measuring device for tools is proposed with the tool identification device, wherein the transmitted-light camera system is designed as at least one part of an optical system for performing a measuring function, in particular a tool measuring function for measuring the tools, the setting and / or measuring device. This allows advantageous properties with regard to process reliability to be achieved when using setting and / or measuring devices.

[0021] Advantageously, the error detection rate of tools in setting and / or measuring devices can be reduced. A "measuring and / or setting device for tools" is understood in particular to mean a device that is intended at least to at least partially detect and / or adjust at least one length, at least one angle, at least one contour, and / or at least one external shape of a tool.

[0022] Furthermore, a computer program product and / or a computer program computing infrastructure is proposed, comprising instructions that, when the computer program is executed by the computing unit, preferably the tool identification device, cause the device to execute the steps of the tool identification method, in particular at least the trained, and in particular further trainable, algorithm, preferably the machine learning algorithm. This allows advantageous properties with regard to process reliability to be achieved when using setting and / or measuring devices.

[0023] Furthermore, a computer-implemented method for training a machine learning algorithm, in particular the trained algorithm used in the tool identification method,proposed. This allows advantageous properties with regard to process reliability when using setting and / or measuring devices to be achieved. If, in at least one preparatory step of the computer-implemented method for teaching the machine learning algorithm, a plurality of different transmitted light images are produced for a plurality of different tool models / tool ​​series of generic tools of a plurality of different tool types, and if, in an algorithm teaching step of the computer-implemented method for teaching the machine learning algorithm, the transmitted light images produced in the preparatory step are combined with the precise tool information associated with the tools contained in the individual transmitted light images, for example, envelope contours determined in advance for the (teaching) transmitted light images,absolute or relative tool positions of the tools depicted in the (teaching) transmitted light images and / or dimensions of the tools depicted in the (teaching) transmitted light images are input into a learning function of the machine learning algorithm, an algorithm optimally trained for the tool identification process, in particular a Kl algorithm, preferably a machine learning algorithm, can advantageously be obtained.

[0024] If, in addition, in a duplication step of the computer-implemented method for training the machine learning algorithm, a plurality of further transmitted light images are artificially generated for at least one, preferably each, transmitted light image of the plurality of transmitted light images recorded in the preparation step, in which the tool contained in the respective transmitted light image is repositioned relative to the originally recorded transmitted light image, wherein in particular in the algorithm training step, the transmitted light images artificially generated in the duplication step together with the transmitted light images produced in the preparation step and together with the exact tool information associated with the tools contained in the individual (artificially generated and actually produced) transmitted light images, for example envelope contours determined in advance for the (training) transmitted light images,If absolute or relative tool positions of the tools depicted in the (teaching) transmitted light images and / or dimensions of the tools depicted in the (teaching) transmitted light images are input into the learning function of the machine learning algorithm, the quality of the trained algorithm, in particular the Kl algorithm, preferably the machine learning algorithm, can advantageously be further increased for use in the tool identification method.

[0025] The tool identification method according to the invention, the tool identification device according to the invention, the setting and / or measuring device according to the invention, the computer program product according to the invention and / or the computer program computing infrastructure according to the invention, and the computer-implemented method according to the invention are not intended to be limited to the application and embodiment described above. In particular, the tool identification method according to the invention, the tool identification device according to the invention, the setting and / or measuring device according to the invention, the computer program product according to the invention and / or the computer program computing infrastructure according to the invention, and the computer-implemented method according to the invention can have a number of individual elements, components, and units that differs from the number stated herein in order to fulfill a functionality described herein.

[0026] Drawings

[0027] Further advantages will become apparent from the following description of the drawings. The drawings illustrate an exemplary embodiment of the invention. The drawings, the description, and the claims contain numerous features in combination. Those skilled in the art will also expediently consider the features individually and combine them into useful further combinations.

[0028] They show:

[0029] Fig. 1 is a schematic perspective view of a setting and / or measuring device for tools with a tool identification device, Fig. 2a is a schematic view of a tool designed as a cutting tool,

[0030] Fig. 2b is a schematic representation of a tool designed as a profiling wheel,

[0031] Fig. 2c is a schematic representation of a tool designed as a profile polishing wheel,

[0032] Fig. 2d is a schematic representation of a tool designed as a form milling cutter,

[0033] Fig. 3 schematic representations of transmitted light images, which were either recorded by means of a transmitted light camera system of the setting and / or measuring device or which were artificially created from a previously recorded transmitted light image,

[0034] Fig. 4 is a schematic flow diagram of a tool identification method for detecting the tools and

[0035] Fig. 5 is a schematic flow diagram of a computer-implemented method for teaching a trained algorithm used in the tool identification method.

[0036] Description of the embodiment

[0037] Figure 1 shows a schematic perspective view of a setting and / or measuring device 32 for tools 10. The setting and / or measuring device 32 has a holding device 44. The holding device 44 is provided for holding a tool 10 or at least one tool chuck 46 receiving a tool 10 (see Fig. 2a). The setting and / or measuring device 32 has an optical system 36. The optical system 36 is provided for performing a measuring function of the setting and / or measuring device 32. The holding device 44 and the optical system 36 can be vertically and / or horizontally adjustable relative to one another. The setting and / or measuring device 32 has a tool identification device 34. The tool identification device 34 comprises a transmitted-light camera system 22. The transmitted-light camera system 22 is formed as part of the optical system 36. The transmitted light camera system 22 has a camera 48.The transmitted-light camera system 22 has an illumination 50. The illumination 50 is arranged opposite the camera 48. A camera field of view of the camera 48 is directed directly onto an illuminated surface of the illumination 50. The transmitted-light camera system 22 is provided for recording transmitted-light images 14 (see Fig. 3) of tools 10. The tool identification device 34 comprises a computing unit 30. The computing unit 30 of the.

[0038] Tool identification device 34 is shown in Fig. 1 as being integrated into the setting and / or measuring device 32, in particular integrally formed with a computing unit 30 of the setting and / or measuring device 32. Alternatively, however, the computing unit 30 could also be designed as an external computing unit.

[0039] The tool identification device 34 is provided for recognizing tools 10. The tool identification device 34 is provided for recognizing tools 10 designed as cutting tools (see Fig. 2a). The tool identification device 34 is provided for recognizing tools 10 designed as profile machining tools (see Figs. 2b and 2c). The tool identification device 34 is provided for recognizing tools 10 designed as form milling cutters (see Fig. 2d).The tool identification device 34 is configured, in particular with the aid of the computing unit 30, to determine at least one tool type of the tool 10, and preferably an exact tool model of the tool 10 and / or an exact tool series of the tool 10, using a trained, and in particular further trainable, algorithm, in particular a K1 algorithm, preferably a machine learning algorithm, at least based on the transmitted-light image 14 recorded by the transmitted-light camera system 22 of a tool 10 in the holding device 44. The computing unit 30 comprises a stored computer program product. The computer program product could also be stored on external data carriers or in a computer program computing infrastructure.The computer program product comprises a computer program with instructions which, when executed by the computing unit 30, cause it to carry out the steps of a tool identification method.

[0040] Figure 2a shows a cutting tool designed as a drill, for example. The drill shown in Figure 2a is clamped in a tool chuck 46, for example. The drill has a cutting edge 52. Figure 2b shows a profile machining tool designed as a profiling wheel, for example. The profiling wheel has a machining surface 54 with a grain. Figure 2c shows a profile machining tool designed as a profile polishing wheel, for example. The profile polishing wheel has a machining surface 54 without grain. Figure 2d shows a form milling cutter designed as a hob, for example. The hob has several cutting edges 52.

[0041] Figure 3 shows exemplary transmitted light images 14, 14', 14", 14"' of the profile processing tool designed as a profile polishing wheel. One of the transmitted light images 14 was recorded directly by the transmitted light camera system 22 and thus represents an original transmitted light image 14. The other transmitted light images 14', 14", 14'" are artificially generated from the recorded original transmitted light image 14.

[0042] The tool identification device 34 is provided for carrying out the tool identification method. Figure 4 shows a schematic flow diagram of the tool identification method. The tool identification method is provided at least for recognizing tools 10 designed as cutting tools, profile machining tools, and / or form milling cutters. In at least one positioning step 20, the tool 10 is positioned in a reproducibly definable, precisely determinable, and / or precisely known relative position to the transmitted-light camera system 22. Alternatively or additionally, in an alternative or additional positioning step 20', the transmitted-light camera system 22 can also be positioned in a reproducibly definable, precisely determinable, and / or precisely known relative position to the tool 10.

[0043] In at least one detection step 12, a transmitted light image 14 of the positioned tool 10 is recorded. The transmitted light image 14 recorded in the detection step 12 comprises at least 40% of a machining contour 24 of the tool 10 in at least one side view of the tool 10. The transmitted light image 14 shown as an example in Fig. 3 comprises 100% of the machining contour 24 of the tool 10 in the side view of the tool 10. The transmitted light image 14 recorded in the detection step 12 can also comprise at least 40% of a machining surface 54 of the tool 10 in at least one side view of the tool 10. The transmitted light image 14 recorded in the detection step 12 is high-resolution.

[0044] In at least one recognition step 16, at least one tool type of the tool 10 is determined by means of the trained, and in particular further trainable, algorithm, in particular by means of a Kl algorithm, preferably by means of a machine learning algorithm, based on the transmitted light image 14 recorded in the recording step 12. In the recognition step 16, a tool model of the tool 10 is also determined by the trained, and in particular further trainable, algorithm based on the transmitted light image 14 recorded in the recording step 12. In the recognition step 16, a tool series of the tool 10 can also be determined by the trained, and in particular further trainable, algorithm based on the transmitted light image 14 recorded in the recording step 12. The trained and / or trainable algorithm is designed as a CNN (Convolutional Neural Network) algorithm.

[0045] In the recognition step 16, the tool type of the tool 10, preferably the exact tool model of the tool 10 and / or the exact tool series of the tool 10, is determined by means of the trained, and in particular further trainable, algorithm based on an envelope contour 18 of the tool 10 determined from the transmitted light image 14 recorded in the acquisition step 12. For this purpose, in at least one sub-step 56 of the recognition step 16, the envelope contour 18 of the tool 10 is extracted from the transmitted light image 14 by an envelope contour determination algorithm of the computing unit 30.In addition, in the recognition step 16, the tool type of the tool 10, preferably the exact tool model of the tool 10 and / or the exact tool series of the tool 10, is determined by means of the trained, and in particular further trainable, algorithm based on an absolute or relative tool position of the tool 10 determined by means of a position recognition algorithm and / or by means of the trained, and in particular further trainable, algorithm from the transmitted light image 14 recorded in the acquisition step 12. For this purpose, in at least one sub-step 58 of the recognition step 16, the relative position of the tool 10 to the view-through camera system 22 is extracted by the position recognition algorithm or by the trained, and in particular further trainable, algorithm.Alternatively or additionally, a (position) parameter of a camera positioning device of the measuring and / or setting device 32 and / or a tool positioning device of the measuring and / or setting device 32 could also be read out to determine the relative position.

[0046] In at least one sub-step 60 of the recognition step 16, based on the transmitted light image 14 recorded in the acquisition step 12, at least the tool type is determined by applying the trained, and in particular further trainable, algorithm. In at least one further sub-step 62 of the recognition step 16, based on the transmitted light image 14 recorded in the acquisition step 12, a dimension of the tool 10 is determined by the position recognition algorithm. In at least one further sub-step 68 of the recognition step 16, based on the transmitted light image 14 recorded in the acquisition step 12, a quadratic roughness of the machining contour 24 and / or the machining surface 54 of the tool 10 or a mean roughness of the machining contour 24 and / or the machining surface 54 of the tool 10 is determined.In at least one further sub-step 64 of recognition step 16, the determined dimension and the determined tool type are combined, and the exact tool model and / or the exact tool series are determined from the combination. In at least one further sub-step 66 of recognition step 16, the determined root mean square roughness or the determined mean roughness and the determined tool type are combined, and the exact tool model and / or the exact tool series are determined from the combination.

[0047] In at least one (optional) output step 26, the transmitted light image 14 of the tool 10 used to carry out the recognition step 16 and / or the determined tool type of the tool 10, preferably the determined tool model of the tool 10 and / or the determined tool series of the tool 10, is output to a user unsolicited or only after prior request by the user.In at least one algorithm training step 28, the feedback from the user receiving the transmitted light image 14 of the tool 10 and / or the determined tool type of the tool 10, preferably the determined tool model of the tool 10 and / or the determined tool series of the tool 10, in the output step 26 regarding the correctness of an identification of the tool type of the tool 10, in particular the exact tool model of the tool 10 and / or the exact tool series of the tool 10, is used for automated optimization and / or for automated training of the trainable algorithm performing the recognition step 16.

[0048] Fig. 5 shows a schematic flow diagram of a computer-implemented method for teaching the trained algorithm used in the tool identification method. In at least one preparation step 38, a plurality of different transmitted light images 14 are produced for a plurality of different tool models and / or tool series of generic tools 10 of a plurality of different tool types. In at least one duplication step 42, a plurality of further transmitted light images 14', 14" (cf. Fig. 3) are then artificially generated for the transmitted light images 14 produced in the preparation step 38. In the artificially generated further transmitted light images 14', 14", 14'", the tool 10 contained in the initial transmitted light image 14 is repositioned relative to a position of the tool 10 in the initial transmitted light image 14.In at least one algorithm training step 40, the transmitted light images 14 created in the preparation step 38 and the transmitted light images 14', 14", 14'" artificially generated in the duplication step 42, together with the precise tool information associated with the tools 10 contained in the individual transmitted light images 14, 14', 14", 14'", are input into a training function of the machine learning algorithm to be trained. Based on this information, the machine learning algorithm is trained and transformed into the trained and, in particular, further trainable algorithm used in the tool identification method of Fig. 4.

[0049] Reference symbol

[0050] 10 tools

[0051] 12 Recording step

[0052] 14 Transmitted light image

[0053] 16 Recognition step

[0054] 18 Envelope contour

[0055] 20 positioning steps

[0056] 22 Transmitted light camera system

[0057] 24 Machining contour

[0058] 26 Output step

[0059] 28 Algorithm training step

[0060] 30 computing units

[0061] 32 Setting and / or measuring device

[0062] 34 Tool identification device

[0063] 36 optical system

[0064] 38 Preparation step

[0065] 40 Algorithm training step

[0066] 42 Replication step

[0067] 44 Holding device

[0068] 46 tool chucks

[0069] 48 Camera

[0070] 50 Lighting

[0071] 52 cutting edge

[0072] 54 Editing interface

[0073] 56 sub-steps

[0074] 58 sub-steps

[0075] 60 sub-steps

[0076] 62 sub-steps

[0077] 64 sub-steps

[0078] 66 sub-step 68 sub-step

Claims

Claims 1. Tool identification method at least for recognizing tools (10) designed as cutting tools, as profile machining tools and / or as form milling cutters, with at least one detection step (12) in which at least one transmitted light image (14) of at least the tool (10) is recorded, and with at least one recognition step (16) in which at least one tool type of the tool (10) is determined by means of a trained, and in particular further trainable, algorithm, in particular a KI algorithm, preferably a machine learning algorithm, at least based on the transmitted light image (14) recorded in the detection step (12).

2. Tool identification method according to claim 1, characterized in that at least one tool model / tool series of tools (10) of identical tool type can be determined by the trained algorithm in the recognition step (16) based on the transmitted light image (14) recorded in the detection step (12).

3. Tool identification method according to claim 1 or 2, characterized in that the trained algorithm is a CNN (Convolutional Neural Network) algorithm.

4. Tool identification method according to one of the preceding claims, characterized in that in the recognition step (16) at least the tool type of the tool (10), preferably an exact tool model / a tool series of a tool (10), is determined by means of the trained algorithm at least based on an envelope contour (18) of the tool (10) determined from the transmitted light image (14) recorded in the detection step (12).

5. Tool identification method according to one of the preceding claims, characterized in that in the recognition step (16) at least the tool type of the tool (10), preferably an exact tool model / an exact tool series of a tool (10), is determined by means of the trained algorithm at least based on a tool position of the tool (10) determined from the transmitted light image (14) recorded in the detection step (12), in particular by means of a position recognition algorithm and / or by means of the trained algorithm.

6. Tool identification method according to claim 5, characterized in that the tool (10) is positioned in a positioning step (20), in particular preceding the detection step (12), in a reproducibly definable, exactly ascertainable and / or exactly known relative position to a transmitted-light camera system (22) recording the transmitted-light image (14).

7. Tool identification method according to claim 5 or 6, characterized in that a transmitted-light camera system (22) recording the transmitted-light image (14) is positioned in an alternative or additional positioning step (20'), in particular preceding the detection step (12), in a reproducibly definable, exactly ascertainable and / or exactly known relative position to the tool (10).

8. Tool identification method according to one of the preceding claims, characterized in that the transmitted light image (14) recorded in the detection step (12) corresponds to at least 40%, preferably at least 60% and preferably 100%, of a machining contour (24) of the tool (10), in particular a machining surface (54) of the tool (10), in at least one side view of the tool (10).

9. Tool identification method according to one of the preceding claims, characterized in that the transmitted light image (14) recorded in the detection step (12) is high-resolution.

10. Tool identification method according to one of the preceding claims, characterized by an output step (26) in which the transmitted light image (14) of the tool (10) used for carrying out the recognition step (16) and / or the determined tool type of the tool (10), preferably the determined Tool model / the determined tool series of the tool (10), is issued to a user unsolicited or only after prior request by the user. 11 . Tool identification method according to claim 10, characterized by an algorithm training step (28), in which a feedback of the transmitted light image (14) of the tool (10) and / or the determined tool type of the tool (10), preferably the determined tool model / the determined tool series of the Tool (10), in the output step (26) receiving user regarding the correctness of an identification of the tool type of the tool (10), in particular an exact tool model / an exact tool series of the tool (10), for an automated optimization and / or for an automated Training of the further trainable algorithm carrying out the recognition step (16), in particular the Kl algorithm, preferably the machine learning algorithm, is used.

12. Tool identification method according to one of the preceding claims, characterized in that the transmitted light image (14) used to carry out the recognition step (16) is recorded from a stationary tool (10).

13. Tool identification device (34), at least for recognizing tools (10) designed as cutting tools, as profile machining tools and / or as form milling cutters, in particular for carrying out a tool identification method according to one of the preceding claims, with at least one transmitted-light camera system (22) for recording at least one transmitted-light image (14) of at least the tool (10), and with at least one internal or external computing unit (30) which is configured to determine at least one tool type of the tool (10), and preferably an exact tool model / an exact tool series of the tool (10), by means of a trained and in particular further trainable algorithm, in particular a KI algorithm, preferably a machine learning algorithm, at least based on the transmitted-light image (14) of the transmitted-light camera system (22).

14. Setting and / or measuring device (32) for tools (10) with a tool identification device (34) according to claim 13, wherein the transmitted-light camera system (22) is designed as at least part of an optical system (36) for carrying out a measuring function of the setting and / or measuring device (32).

15. Computer program product and / or computer program computing infrastructure, comprising instructions which, when the computer program is executed by a computing unit (30), preferably a tool identification device (34) according to claim 13, cause the latter to execute the steps of the tool identification method according to one of claims 1 to 12, in particular at least the trained algorithm, preferably the machine learning algorithm.

16. Computer-implemented method for training a machine learning algorithm, in particular the trained algorithm used in the tool identification method according to one of claims 1 to 12, characterized in that in at least one preparation step (38) a plurality of different transmitted light images (14) are produced for a plurality of different tool models / tool ​​series of generic tools (10) of a plurality of different tool categories, and in that in an algorithm training step (40) the transmitted light images (14) produced in the preparation step (38) together with the exact tool information associated with the tools (10) contained in the individual transmitted light images (14) are entered into a training function of the machine learning algorithm.

17. Computer-implemented method according to claim 16, characterized in that in a duplication step (42) for at least one, preferably each, transmitted light image (14) of the plurality of transmitted light images (14) recorded in the preparation step (38), a plurality of further transmitted light images (14', 14", 14'") are artificially generated, in which the tool (10) contained in the respective transmitted light image (14) is repositioned relative to the originally recorded transmitted light image (14), wherein in particular in the algorithm training step (40) the transmitted light images (14', 14", 14'") artificially generated in the duplication step (42) are combined with the transmitted light images (14) produced in the preparation step (38) and together with the respective (artificially generated and actually produced) transmitted light images (14, 14', 14",14'”) contained tools (10) are entered into the learning function of the machine learning algorithm.,

Images